IZA DP No. 1344 Input versus Output Taxation in an Experimental International Economy Arno Riedl Frans van Winden DISCUSSION PAPER SERIES Forschungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor October 2004
IZA DP No. 1344
Input versus Output Taxation in anExperimental International Economy
Arno RiedlFrans van Winden
DI
SC
US
SI
ON
PA
PE
R S
ER
IE
S
Forschungsinstitutzur Zukunft der ArbeitInstitute for the Studyof Labor
October 2004
Input versus Output Taxation in an Experimental International Economy
Arno Riedl CREED, University of Amsterdam, Tinbergen Institute and IZA Bonn
Frans van Winden
CREED, University of Amsterdam and Tinbergen Institute
Discussion Paper No. 1344 October 2004
IZA
P.O. Box 7240 53072 Bonn
Germany
Phone: +49-228-3894-0 Fax: +49-228-3894-180
Email: [email protected]
Any opinions expressed here are those of the author(s) and not those of the institute. Research disseminated by IZA may include views on policy, but the institute itself takes no institutional policy positions. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit company supported by Deutsche Post World Net. The center is associated with the University of Bonn and offers a stimulating research environment through its research networks, research support, and visitors and doctoral programs. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.
IZA Discussion Paper No. 1344 October 2004
ABSTRACT
Input versus Output Taxation in an Experimental International Economy∗
This paper is concerned with a policy oriented macroeconomic experiment involving an 'international' economy with a relatively small 'home' country and a large 'foreign' country. It compares the economic performance of two alternative tax systems as a means to finance unemployment benefits: a sales-tax-cum-labor-subsidy system versus a wage tax system. The two systems are applied to the home country, while the wage tax system always obtains in the foreign country. In stark contrast with expectations of experts the sales tax system clearly outperforms the wage tax system, using standard economic indicators. It is argued that producers' reluctance to incur costs up-front while being uncertain about product prices can explain this outcome. Several pieces of evidence are provided to support this claim. The results strongly suggest that behavioral aspects have to be taken into account also in applied macroeconomic models. JEL Classification: A10, C90, C91, D21, D80, E62, H20 Keywords: laboratory experiment, wage tax, sales tax, macroeconomic policy, behavioral
economics Corresponding author: Arno Riedl CREED University of Amsterdam Roetersstraat 11 1018 WB Amsterdam The Netherlands Email: [email protected]
∗ This paper uses data generated by a research project commissioned by the Dutch Ministry of Social Affairs and Employment, see van Winden, Riedl, Wit, and van Dijk (1999). Financial support by the Ministry and comments and suggestions by its Steering Committee are gratefully acknowledged. We would like to thank Frans van Dijk and Jörgen Wit who were both involved in the experimental design of the project, and, in addition, Jörgen Wit for the computation of the equilibria of the theoretical model used. Our gratitude also goes to Jos Theelen for the development of the excellent software used in the experiment, and to G. Cotteleer, J.H.H. Notmijer, and M. Smits for their assistance in running the experiments. Former versions of this paper have been presented at conferences, workshops and seminars in Amsterdam, Barcelona, Bari, Berlin, Bonn, Frankfurt, Groningen, Lake Tahoe, Munich, New York, Rhodos, St. Gallen, The Hague, Tilburg, Venice, Zurich. We thank the participants for their helpful remarks. We are in particular grateful to Kurt Hildenbrand, Ruud de Mooij, Charles Noussair, Raymond Riezman, Shyam Sunder, and Claus Weddepohl for their valuable comments. This paper is part of the EU-TMR Research Network ENDEAR (FMRX-CT-98-0238). The usual disclaimer applies.
1 Introduction
Time usually elapses (...) between the incurring of costs by the producer
(with the consumer in view) and the purchase of the output by the ultimate
consumer. Meanwhile the entrepreneur (...) has to form the best expecta-
tions he can as to what the consumers will be prepared to pay when he is
ready to supply them (...).
John Maynard Keynes (1970 [1936], Ch. 5: Expectation as Determining
Output and Employment, p. 46)
A major economic issue concerns the effects of taxation on the behavior of individual
consumers and producers and the performance of markets. In this context, a longstand-
ing problem in public finance relates to the pros and cons of taxing inputs, e.g. labor
and capital, versus the taxation of outputs, like sales or value added. One potentially
highly relevant factor in this respect is that production takes time, a fact emphasized by
Keynes in the preceding quote. At the time when producers have to make their input
decisions, generally, the precise market conditions prevailing at the time consumers buy
their products are unknown. Thus, when deciding on labor and capital employment
producers, typically, face uncertainty about the real returns from these decisions. A
similar problem holds for consumers when they have to allocate time between labor
and leisure, because the real return on their labor will depend on the development of
consumer prices over the period covered by the wage contract.
Several studies have argued that taking this uncertainty into account is important
from a behavioral explanatory and optimal policy point of view. For example, Eaton
and Rosen (1980) show that if consumers are uncertain about the real wage, an ex-
pected income-compensated increase in the wage tax may induce them to supply more
labor. Moreover, lump-sum taxation is no longer necessarily efficient, because the wage
tax insures the consumer against random real wage income movements. Regarding pro-
ducers, a number of theoretical partial equilibrium studies have focused on the effects
of output price uncertainty on the input and supply decisions of firms. Results show
that output price uncertainty generally reduces factor demand and production level
of risk-averse competitive firms (Sandmo 1971, Batra and Ullah 1974, Hartman 1975,
1976, Holthausen 1976, Ghosal 1995).1
1Loss aversion, as in prospect theory (Kahneman and Tversky 1979), would seem to make this effect
1
The policy relevance of this topic can be illustrated by referring to “the puzzle of
European unemployment” (Blanchard and Katz (1997)). A large piece of this puzzle
seems related to the strong reliance on wage taxation in financing the welfare state, and
the focus on supply side conditions in employment policies. Indeed, several scholars
have pointed at the pernicious effects of wage taxation in this respect, with rising tax
rates and unemployment leading to a vicious circle (Snower (2000)). These arguments
and the above mentioned research suggests that shifting taxation from inputs to outputs
may have a positive effect on production and employment because the government then
effectively shares the sales risk faced by producers.
From an optimal taxation and general equilibrium perspective, however, it seems
not at all clear whether such a shift in taxation will do any good, in particular in a small
open economy. Taxation of outputs implies an implicit tax on the mobile factor capital
and the conventional wisdom in the literature on optimal taxation in open economies
is that taxing such a factor should be avoided. For example, based on the seminal work
of Diamond and Mirrlees (1971), Razin and Sadka (1991) have shown that a small
open economy should not tax mobile capital (at the source). More recently, Bovenberg
(1994, p. 284) argued that “(...) small and open economies should not tax highly mobile
factors (...)”.2
On the other hand, pure theoretical reasoning does not always provide unambiguous
answers, even in a frictionless perfectly competitive world. A well-known result from
general equilibrium theory is that, generically, equilibrium predictions are not unique.
In case of multiple equilibria, however, no clear forecasts concerning policy reforms can
be made. A main motivation of this paper is, therefore, to shed some light on this
thorny issue of whether a tax on immobile labor or a sales taxes - implicitly taxing
mobile capital - leads to a better economic performance of a small open economy. In
fact, in the framework we will be using theoretically multiple equilibria arise.
only stronger. Another strand of literature addresses the impact of (macroeconomic) uncertainty on
investment, typically showing a negative effect (Aizenman and Marion 1993, Brunetti and Weder 1998,
Guiso and Parigi 1999).
2Many relatively small countries nevertheless tax capital implicitly or explicitly. A large body of
literature tries to square this empirical fact with the theory of optimal taxation either by discussing
legal details (e.g. Gordon 1992), allowing for frictions and market imperfections (e.g. Richter and
Schneider 2001, Koskela and Schob 2002) or taking a global view of capital taxation (Braulke and
Corneo 2003).
2
Another motivation relates to the novelty of the research method. For our inves-
tigation we use data from an experimental study pitting a wage tax system against a
sales tax system as alternative means to finance unemployment benefits, commissioned
by the Dutch Ministry of Social Affairs and Employment.3 The Minister was requested
to do so in a motion carried by the Second Chamber of the Dutch parliament. To
our knowledge, it is for the first time that policymakers explicitly asked for laboratory
experimentation as a means to advise in macroeconomic policymaking. When doing
our investigation we were supervised by a steering committee to which internationally
renowned Dutch economists (in the fields of public economics, labor economics, exper-
imental economics and applied general equilibrium modeling) were assigned.4 Being a
policy-oriented study, the experimental design was required to show some parallelism
with the Dutch economy. The steering committee had to approve the design of the
experiment and assist the project.
Further innovative aspects of our study concern the comparison of different tax
systems in a macroeconomic experiment, and the implementation of a relatively small
‘home’ economy and a large ‘foreign’ economy in the laboratory.5 In a sense, doing
this study meant exploring the boundaries of the research method of laboratory ex-
perimentation. In our view, the results show that also in this area of policy related
macroeconomic research experiments are a useful complementary research tool, next
to theoretical and field empirical analyses. Compared to field econometric studies an
important advantage is that it is possible to empirically analyze the economic conse-
quences of a complete implementation of a new tax system. With the additional virtue
of being able to do so in a controlled way. Furthermore, an experiment offers the op-
portunity to generate (and if necessary replicate) the micro-level data of interest and
3See van Winden, Riedl, Wit, and van Dijk (1999).
4For economically intuitive reasons, backed by the above mentioned theoretical results from optimal
taxation theory, the members of the committee had the general opinion that the sales tax system
would lead to capital flight, more unemployment, and a substantial welfare loss in a relatively small
open economy, like The Netherlands. In addition, it was feared that a shift in economic activity would
take place from the relatively capital intensive ‘exposed sector’ (producing tradeable goods) towards
the more labor intensive ‘sheltered sector’. The more so, because high tax rates were foreseen due to
a labor subsidy that was incorporated in the alternative sales tax system.
5Akerlof (2002) discusses some other recent experiments related to macroeconomic issues.
3
avoids the noise field data are unavoidably exposed to.6 In addition, no specific behav-
ioral assumptions are needed, nor a restriction to a partial equilibrium framework as
in the theoretical studies referred to above. Moreover, since theory generically predicts
multiple equilibria, experiments can provide information on their relative attractiveness
in practice; an issue that will also prominently show up in our study.
More specifically, the experimental international economy that we will investigate
consists of two ‘countries’, one of which - the home country - is relatively small in terms
of potential economic activity. In each country consumers and producers are active.
Consumers supply labor and capital to producers on local and global input markets.
In both countries, producers are distributed over two production sectors: a sheltered
sector producing a relatively labor intensive commodity for a local output market, and
an exposed sector producing a relatively capital intensive commodity for a global output
market. All production factors and consumption goods are traded through multi-unit
double auctions.7 In the benchmark experimental treatment, in both countries, a wage
tax finances the benefits consumers receive for unemployed labor. In the alternative
treatment, the wage tax system is substituted by a sales tax system, in the home
country only. Under this system, instead of having to pay a tax on labor up-front, a
producer is taxed according to the proceeds from sales. Moreover, for each employed
unit of labor the producer receives a subsidy equal to the unemployment benefit.
The theoretical general equilibrium predictions turn out to be unique for the wage
tax system. For the alternative sales-tax-cum-labor-subsidy system, however, we obtain
two stable general equilibria implying two quite distinct sets of theoretical predictions
concerning the economic performance indicators of this system. One set of predictions
supports the economically intuitive hypothesis of capital flight from the small to the
large country with very negative effects on factor employment, production, consumption
and, hence, welfare in the small country. In the second equilibrium, however, almost
no capital flight occurs and labor employment, production, and consumption levels are
even higher than under the benchmark system. The experiment allows us to investigate
whether economic activities are attracted to one of these equilibria.
6In empirical studies of taxation this is a notorious problem which, for example, manifests itself in
widely diverging estimates of tax rate elasticities (see e.g. Sørensen (1997)).
7Double auctions are typically used for their capability to facilitate the equilibration of supply and
demand and the generation of efficient outcomes (see e.g. Davis and Holt (1993)).
4
To evaluate the performance of the two tax systems relative to each other as well as
relative to the theoretical predictions we mainly use the following economic indicators:
employment of labor, capital flight, shift towards labor intensive production, real GDP,
consumer earnings, and the budget surplus. Our main findings are the following. First,
despite of the rather complex experimental environment, we observe a clear tendency
towards equilibration of the economic process. Second, it turns out that the wage
tax system shows persistent budget deficits, while tax adjustments to balance these
deficits have a strong negative impact on the employment of labor and real GDP.
Third, shifting taxation from wages to sales and subsidizing labor in the home country
has substantial positive budgetary and real economic effects for this country. Moreover,
there is no evidence of capital flight nor of a shift in economic activity towards the labor
intensive sector. Fourth, under the alternative sales tax system economic behavior
tends to coordinate on the equilibrium with the higher activity level or performs even
better than this equilibrium predicts. In summary, the alternative sales-tax-cum-labor-
subsidy system performs significantly better than the wage tax system.
To explain these findings we claim that producers’ aversion towards incurring costs
up-front, while facing output price uncertainty, plays a crucial role. The sales-tax-cum-
labor-subsidy system is clearly much more producer and employment friendly in this
respect. Instead of having to pay a tax on the input of labor, a subsidy is received, while
through the sales tax the government is sharing the risk the producer runs with respect
to the return on output. We present theoretical arguments and empirical evidence to
support our claim.
Our results point at a hitherto underexposed behavioral regularity, with relevance
for economic model building as well as policy advising. Regarding the latter our study
fits into a still small but gradually growing stream of ‘design’ studies which involve the
economist as ‘engineer’ (Roth 2002). In these studies experimental and computational
economics are used as research methods filling the gap between theory and design.
For the development of theory these studies can be helpful by posing challenges and
suggesting some new answers to open questions. However, as Roth notes: “Whether
economists will often be in a position to give highly practical advice on design depends
in part on whether we report what we learn, and what we do, in sufficient detail to allow
scientific knowledge about design to accumulate” (ibid., p. 1342). With our paper we
hope to make a contribution to this empirical feedback mechanism.
5
The organization of the paper is as follows. Section 2 presents the experimental
design and procedures, as well as the theoretical predictions. The experimental results
are given in Section 3. In Section 4 we propose a behavioral explanation for our main
findings, while additional supportive evidence is provided. Section 5 concludes.
2 Experimental design
In the following, the wage tax system is denoted as the WT-system and the alternative
sales-tax-cum-labor-subsidy system as the STLS-system.
2.1 Economic Environment
In view of the desired parallelism with a relatively small open economy, we consider
an ‘international’ economy with consumers and producers in two ‘countries’, a rela-
tively small country s, the home country, and a large country l, the foreign country.
Consumers are endowed with units of capital (K) and labor (L) that they can sell to
producers in a capital and a labor market. Consumers derive utility from ‘leisure’, i.e.
unsold units of labor, and the consumption of two private goods: X and Y . In addi-
tion to factor payments, the consumption budget is determined by an unemployment
benefit for each unsold unit of labor. Commodities X and Y are produced in separate
sectors. Producers need capital and labor as inputs, which are transformed to outputs
via CES production technologies. The production of good X is relatively capital in-
tensive, while the production of Y is relative labor intensive. Profits are determined by
the difference between sales revenue and the costs of inputs. The former may involve
sales taxes and the latter wage taxes or labor subsidies, depending on the prevailing
tax system. Taxes are paid for the finance of unemployment benefits and/or labor
subsidies (see the next subsection). Both the capital market and the market for X are
international (exposed), while the markets for labor and good Y are local (sheltered).
Consequently, the total number of input and output markets equals six. Figure 1 shows
a flow diagram illustrating the economic environment.
Consumers are endowed with K units of capital and L units of labor. Preferences
over leisure (L − L) and the two consumption goods, X and Y , are induced by a log-
6
Consumers insmall country
Producers Y insmall country
Producers X insmall country
Producers X inlarge country
Producers Y inlarge country
Consumers inlarge country
'&
$%
Labor market insmall country
'&
$%
Market for Y insmall country
'&
$%
Internationalmarket for X
'&
$%
Market for Y inlarge country
'&
$%
Labor market inlarge country
'&
$%
Internationalcapital market
- - -
- - -
@@
@@
@@
@R
��
��
��
��
QQQs
��
��
���3
����1
PPPPq
��
��
��
��
��3
QQs
ZZ
ZZ~
��
��>
?
6
?
6
Figure 1 – Flow diagram of the economic environment
linearized Cobb-Douglas type of utility function.8 Producers are endowed with a CES
production technology exhibiting slightly decreasing returns to scale and allowing for
different factor intensities and elasticities of substitution in the two production sectors.9
In the upper part of Table 1 the continuous approximations of the discrete utility
(earnings) and output tables used in the experiments are shown. The rest of this table
will be discussed below.
All inputs and outputs are traded in computerized multiple units double auction
markets as introduced by Plott and Gray (1990). The choice of this market type is
guided by its reputation of fast equilibration of supply and demand in experimental
8The use of a log-linearized Cobb-Douglas utility function has the advantage that subjects could
be provided with a simple sheet of paper showing the marginal and total payoff for each of the three
arguments, even though three goods entered the utility function as variables.
9The actually implemented factor intensities and substitution elasticities resemble estimates for the
Dutch economy. The choice of slightly decreasing returns to scale is motivated by an empirical and
a methodological consideration. Firstly, empirical evidence exist supporting this choice (see Basu and
Fernald (1997)). Secondly, it allows experimental producers to make strictly positive profits, and hence
monetary earnings, in the theoretical general equilibrium discussed below.
7
Table 1 – Experimental parameters
Preferences and production technologies
Consumers i (utility functions):
Uik= 25
[
ln Xik+ ln Yik
+ .25 ln(Lik− Lik
)]
,
Uik= 0 if either Xik
, Yik, or Lik
− Likequals zero,
}
in both tax systems (k = s, l)
Quantities Lik, Yik
are determined ‘locally’ (within a country)
Quantities Xikare determined ‘internationally’ (one global market)
Producers j (production functions and profit functions):
Zjzk= Ak
[
η1−γzz Lγz
jzk+ (1 − ηz)
1−γz Kγz
jzk
] 0.9γz , Z = X, Y ; z = x, y; k = s, l
Labor intensities : ηx = .5625, ηy = .675; Substitution elasticities : γx = −2, γy = −6
Scaling factor: As = 1 (small country), Al = 1.21 (large country)
Πjxk= pxXjxk
− (1 + τwk)wkLjxk− rKjxk
,
Πjyk= pykYjyk
− (1 + τwk)wkLjyk− rKjyk
, k = s, l
}
in WT-system
Πjxs = (1 − τxs)pxXjxs − (ws − w0)Ljxs − rKjxs ,
Πjys = (1 − τys)pysYjys − (ws − w0)Ljys − rKjys ,
Πjxl= pxXjxl
− (1 + τwl)wlLjxl− rKjxl
,
Πjyl= pylYjyl
− (1 + τwl)wlLjyl− rKjyl
,
in STLS-system
Prices pyk, wk, taxes τwk, τzs, and quantities Ljzk, Yjyk
are determined ‘locally’
(within country k = s, l)
Prices px, r, and quantities Kjxk, Xjxk
are determined ‘internationally’
(one global market)
Endowments (both tax systems)
Small country Large country
Consumer Li = 15, Ki = 10, Cashi = 181 Li = 105, Ki = 70, Cashi = 1268
X-producer Lj = 0, Kj = 0, Cashj = 1223 Lj = 0, Kj = 0, Cashj = 8557
Y -producer Lj = 0, Kj = 0, Cashj = 815 Lj = 0, Kj = 0, Cashj = 5705
Number of agents
Consumers 3 3
X-Producers 2 2
Y -Producers 3 3
Tax systems
WT-system STLS-system
Both countries k Small country s Large country l
Unemployment benefit (w0) 70 70 70
Labor subsidy (w0) 0 70 0
Initial wage tax rate (τ0w) .3777 0 .3777
Wage tax τ t+1
wk wtkLt
k = τ t+1
wl wtkLt
l =
adjustment rule (τ t+1w ) w0(Lk − Lt
k) w0(Ll − Ltl)
Initial sales tax rate X (τ 0x) 0 .6521 0
Initial sales tax rate Y (τ0y ) 0 .7518 0
Sales taxes τ t+1xs pt
xXts + τ t+1
ys ptysY
ts = w0Ls
adjustment rule (τ t+1x , τ t+1
y ) τ t+1xs /τ t+1
ys = τ0xs/τ0
ys
Note: In the table describing the tax systems, t denotes a trading period, the variables Ltk, Lk, Xs, and Ys denote
aggregates in a country, superscripts 0 refer to initial values.
8
market economies. Trading takes place in a number of trading periods. Each trading
period is split into a first phase with only the input markets open, and a second phase
with only the output markets open.10 To facilitate trading, both consumers and pro-
ducers are endowed with some fiat money (Cash in Table 1) at the beginning of the
first phase of each period. In addition, consumers receive a transfer (unemployment
benefit w0) for each unit of labor that is unemployed at the end of this phase.11
All taxes are levied on the producers. In the baseline treatment of the experiment
the WT-system obtains in both countries. In this case a given tax rate (τwk, k=s, l) is
applied to the wage of each unit of labor that is employed. In the treatment concerning
the alternative tax system the WT-system again obtains in the large country, but now
the STLS-system prevails in the small (home) country. Instead of paying a wage tax,
producers in the small country now receive a fixed subsidy (equal to the unemployment
benefit) for each unit of labor they employ, while paying a given tax rate (τxs in the
X-sector and τys in the Y -sector ) on the sales price of their products.12
Experimental subjects participate in a sequence of 16 trading periods. In a given
treatment the first eight of these periods are identical with respect to the exogenous
parameters. Except for the subjects’ earnings nothing carries over from period to pe-
riod. Consequently, each period can be seen as a repetition of the same static economy.
10There are two main reasons for using sequential instead of simultaneous markets. Firstly, in our
view sequentiality of input and output markets is much more common in the field than simultaneous
markets are. Note also that even simultaneous markets would exhibit some sequentiality, were it
alone for the sequentiality that is inherent to the production process (cf. Keynes’ view quoted in the
beginning). Secondly, the sequentiality considerably reduces the complexity of the market environment
for the subjects.
11Any unemployment remaining in an equilibrium may be viewed as ‘voluntary’, theoretically (ac-
cording to Layard, Nickell, and Jackman (1991, p. 41), however, the question of voluntary versus
involuntary unemployment is ‘fruitless’ for practical and public policy purposes). Rogerson (1997)
convincingly argues that any distinction between voluntary and involuntary, frictional and cyclical,
equilibrium and disequilibrium unemployment is meaningless because all unemployment consists of all
these components. The inclusion of ‘frictions’, like trade unions or efficiency wages, might have added
some realism. However, in view of the novelty of the set-up and the already complicated nature of the
economy it was decided to start with a relative simple market structure. Lian and Plott (1998) use a
similar setup for their general equilibrium experiment.
12Because this study does not focus on transitional issues, a between-subjects design was chosen for
the tax systems.
9
In periods 9-16 tax rates are adjusted at the beginning of each new period such that
a balanced budget would be obtained for the previous period, given the market out-
comes of that period. The initial tax rates and the precise tax adjustment rules are
shown in the lower part of Table 1.13 This procedure guarantees a sufficient number
of repetitions with a constant environment for making it possible to examine whether
and at which level economic behavior stabilizes. The adjustment of the tax rates to
the budget balance adds an important feature of realism and enables an analysis of
the dynamic interaction between taxation, employment and other indicators of eco-
nomic performance, while keeping everything else constant. It also allows to control
for the potentially confounding effect that a relative good performance of a tax system
is ‘bought’ by budget deficits.
Table 1 shows the parameter values chosen for the endowments, utility functions,
production functions, and the number of agents. To implement a large country in the
laboratory the following solution was chosen. While keeping the number of consumers
and producers the same for both countries, consumers in the large country are endowed
with seven times as many units of labor and capital as holds for the consumers in the
small country (see the different L and K in the table). Moreover, the scaling factors
(As and Al) in the production functions are adjusted such that, theoretically, supply
and demand in the large economy are seven times as large as in the small economy, in
the baseline treatment with the WT-system (see next subsection).14
2.2 Theoretical General Equilibrium Predictions
Given the complex nature of the experimental economy, with several interdependent
markets and the double auction trading mechanism, the most natural solution concept
is the general equilibrium. We calculated the numerical solution(s) of a competitive
general equilibrium model equating supply and demand in the various markets with
13An upper bound of 0.90 was maintained for the tax rates because pilot studies showed that tax
rates too close to 100% might have a strongly discouraging effect on trading.
14The alternative approach of increasing the number of agents instead of endowments would not
have been feasible. With the requirement of at least three agents on each side of a market to ensure
competitiveness (see Davis and Holt (1993), Huck, Konrad, Muller, and Normann (2001)), the minimal
number of subjects per experimental session would have been 64, exceeding by far the capacity of the
laboratory.
10
the additional requirement of a balanced tax-transfer budget. We thereby follow other
studies of experimental markets using a similar procedure (see e.g. Noussair, Plott,
and Riezman (1995, 1997), Quirmbach, Swenson, and Vines (1996)).
Table 2 shows the predictions concerning quantities, relative prices and taxes sep-
arately for the international markets, the small country, and the large country. The
most interesting part of these predictions is the fact that there exist two (stable) equi-
libria for the STLS-system.15 These two equilibria are quite different, in particular for
the small country. Equilibrium 2 shows the serious negative economic consequences
- including a substantial capital flight - for the small country that economic intuition
and the literature on optimal (capital) taxation in small open economies suggests (see
e.g. Diamond and Mirrlees (1971), Bovenberg (1994)). In contrast to the WT-system,
where only labor is taxed, the STLS-system implicitly taxes capital thereby reducing
the rents from capital in the small country. This induces the capital flight observed in
equilibrium 2.
Yet, implicitly taxing capital and labor broadens the tax basis, which has potentially
positive efficiency effects. In particular, since the production process exhibits decreasing
returns to scale allowing tax shifting to an immobile third factor. This is akin to the tax
shifting effect known in the literature on the ‘double dividend’ of environmental taxes
(e.g. de Mooij and Bovenberg (1998). This potentially beneficial tax shifting offers
a rationale for equilibrium 1 in the STLS-system. This equilibrium shows substantial
positive employment effects, little capital flight, and an increase in the production in
both sectors.
The existence of two quite distinct general equilibria for the alternative STLS-
system makes the experimental investigation particularly interesting. In addition to
the comparison of the two alternative tax-transfer systems we can also investigate
whether actual behavior converges to one of the equilibria, if it converges at all. This
is of special interest because the multiplicity of equilibria leaves the economies with a
coordination problem and the theoretical prediction ambiguous.
15Actually, the WT-system also exhibits two (stable) equilibria. Accidentally, however, they are so
close to be virtually not distinguishable. Therefore, we report only one here. Although, generically,
an odd number of general equilibria exist (Dierker (1972)), instable equilibria are neither likely to be
detected - which explains the even number found - nor of particular interest, here.
11
Table 2 – Theoretical general equilibrium predictions
WT-system STLS-system
equilibrium 1 equilibrium 2
International
K 240 240 240
X 177 181 182
r 0.0307 0.0295 0.0289
px 0.1882 0.1807 0.1807
Small country
Ks 30 28 11
Ls 28 33 18
Xcs 22 25 17
Xps 22 25 14
Ys 19 21 11
ws 0.1694 0.1971 0.1292
pys 0.2211 0.2165 0.2747
τws 0.3777
τxs 0.4889 0.7835
τys 0.5414 0.8677
Large country
Kl 210 212 229
Ll 197 199 213
Xcl 155 156 165
Xpl 155 156 168
Yl 132 133 140
wl 0.1694 0.1640 0.1743
pyl 0.2211 0.2123 0.2121
τwl 0.3777 0.3655 0.2769
Note: Equilibrium quantities are rounded to integers. De-
picted prices are relative prices that are obtained by divid-
ing nominal prices by the sum of all six nominal prices. The
equilibrium tax rates guarantee a balanced budget in equi-
librium. Superscript c (p) indicates consumed (produced)
quantities; when this distinction is not made consumed and
produced quantities coincide in equilibrium.
In order to avoid a potential bias of the experimental results in favor of the alter-
native tax system, and because the experiment was also policy oriented, it was decided
not to take the initial tax rates for the STLS-system from one of the two equilibria of
the theoretical model. Instead, these were determined such that on impact the produc-
ers of X and Y would have to bear the same tax burden as empirically observed (in
12
the laboratory) under the WT-system.16 Interestingly, it turned out that these empir-
ically based tax rates (τxs and τys) were in a close neighborhood of the theoretical tax
rates in the ‘bad’ equilibrium 2 prediction. Finally, it is noted that only one currency
(‘francs’, with a fixed conversion rate to Dutch guilders) is used in the lab economy.
Since the focus of this study is not on issues of international finance we did not want
to complicate the experiment by introducing multiple currencies.
2.3 Research Questions and Experimental Procedures
In the following we empirically analyze the performance of the WT- and the STLS-
system in comparison to the theoretical predictions and also compare the two tax sys-
tems with each other. We do this on the basis of relative prices and quantities produced
and consumed. Additionally, with an eye on the political relevance of our study, we
also investigate the behavior of the following more aggregate performance indicators:
employment of labor, net capital export, shift towards labor intensive production, real
GDP, consumer earnings, and the budget surplus. In light of the implicit taxation of
capital under the STLS-system, with capital being mobile and labor immobile between
countries, an intuitive hypothesis predicts that serious capital flight with bad economic
consequences will be observed. This in turn will have negative consequences on labor
employment, real GDP, consumer earnings, and budget surplus. Additionally, an in-
crease in labor intensity of production under the STLS system can be expected. This
was in fact the view taken by our naıve principals.
All experimental sessions were run at the CREED-laboratory of the University
of Amsterdam. Subjects, recruited through announcements on bulletin boards, were
undergraduates of the University and mostly coming from its Faculty of Economics and
Econometrics. Because of the relative complexity of the experimental environment we
implemented the following procedure. Subjects had to sign up for three meetings: a
training session (where participants got acquainted with the trading rules, forms and
tables to be used, and how to handle the computer), a ‘closed economy’ session (for
16More precisely, the initial wage tax rate τ 0w which ceteris paribus balances actual average tax
revenue with actual average unemployment expenditure in the periods 6-8 of the WT-system (denoted
by A) is derived from: τ0wwA
s LAs = w0(Ls − LA
s ). The initial tax rates of the STLS-system (τ 0x and τ0
y )
then follow from: τ0xpA
x XAs − w0L
Axs = τ0
wwAs LA
xs and τ0y pA
ysYA
s − w0LAys = τ0
wwAs LA
ys. When the tax
rates are adjusted, in periods 9-16, the ratio of the tax rates is kept the same (see the lower part of
Table 1).
13
getting subjects experienced with trading), and the international economy session.17
Subjects were paid out only at the end of the third meeting. They received a show-up
fee of 70 Dutch guilders for the training session. In the closed economy sessions they
earned on average 27 guilders, while receiving 40 guilders as a show-up fee. The show-up
fee for the international economy session was 10 guilders, while average earnings in this
sessions amounted to 120 guilders (at the time of the experiments one Dutch guilder
was worth approximately 0.52 U.S. dollar). All meetings lasted about 3.5 hours. At the
training session each subject was randomly assigned the role of consumer or producer,
which they kept in the subsequent meetings.
At the beginning of an experimental session subjects received instructions consist-
ing of a general part, read aloud by the experimenter, and a role-specific part, which
was quietly read by the subjects. They further received personal history forms with
all the information that was relevant to them (concerning endowments, markets they
were allowed to trade in, any taxes or subsidies, and the conversion rate of ‘francs’
to guilders).18 Similar information was provided on the computer screen. By having
them fill in their transactions and earnings these forms were also intended to make
subjects fully aware of the consequences of their decisions. Quizzes were used to check
the understanding of the procedures, the reading of the table with redemption val-
ues (‘utility’) or input-output combinations (production schedule), and the calculation
of earnings. A sample copy of the instructions, trading rules, and personal forms
used in the experiments can be downloaded from http://www1.fee.uva.nl/creed/pdf
files/instr2taxsyscomp.pdf.
Each experimental session started with two unpaid practice rounds, followed by 16
trading periods. During the first eight periods tax rates were kept at their initial values.
From trading period 9 on, they adjusted to the budget balance of the previous period.
In each period, the input markets phase lasted 4 minutes and 30 seconds. Then, after
a short break of 20 seconds, the output markets phase started which lasted 3 minutes
17Parameter values of the closed economy were similar but not identical to the ones used in the exper-
iment. Subjects were selected for the international economy session on the basis of their performance
(earnings) in the closed economy session; they got informed about this at the first meeting.18In the experiment consumers were labeled ‘type-1 traders’ and producers ‘type-2 traders’. More-
over, labor and capital were denoted as good V and good W, respectively. Markets were labeled as
V1(2), W1, X1, Y1(2). The unemployment benefit was denoted as a subsidy for unsold units of V.
14
Table 3 – Summary of experiments
Number of Tax system Number of Number of
subjects in small country periods† constant tax periods
session 1 16 WT 16 (2) 8
session 2 16 WT 16 (2) 8
session 3 16 WT 16 (2) 8
session 4 16 STLS 16 (2) 8
session 5 16 STLS 16 (2) 8
session 6 16 STLS 16 (2) 8
Note:† number of practice periods in parentheses.
and 30 seconds. This was followed by a 2 minutes break for recording before the next
period began.19
Two series of experimental sessions were conducted, each consisting of three sessions.
One series concerned the treatment where the WT-system obtained in both countries,
while the other series dealt with the treatment where the STLS-system was effective in
the small country while the WT-system again prevailed in the large country. Table 3
characterizes the sessions.
3 Experimental Results
In presenting our results we will focus first on the trading periods with a constant tax
regime (periods 1-8). In the constant tax regime of the WT-system the tax rates are
set at the level of the theoretical predictions shown in Table 2. We use the results of
these periods for a comparison with these predictions.20 Yet, the main focus of our
analysis will be on the economic indicators showing the relative performance of the two
tax systems. Recall that in the large country the wage tax system is effective in both
experimental treatments, the WT-system and the STLS-system.
19Standing bids and asks were presented as ‘market prices’ (excluding any taxes or subsidies) and as
‘inclusive prices’ (including taxes or subsidies). After the closing of the factor markets consumers were
informed about the transfers received for unsold units of labor, while producers were informed about
the number of goods produced with the inputs they bought. In addition, some market statistics were
provided concerning trades, average prices, and the average price subjects received (paid) for the inputs
they sold (bought). Similar market statistics were provided after the closing of the product markets.20Recall that the initial tax rates in the STLS-system are determined by using the outcomes of the
constant tax regime of the WT-system.
15
3.1 Constant Tax Regime
Figures 2 and 3 illustrate the development of quantities (panels (a)) and relative prices
(panels (b)), averaged over sessions, for the WT-system and the STLS-system. In
this subsection we restrict our discussion to the left-hand part of each figure (the first 8
periods). The figures show an orderly development, as is also observed in other multiple
markets experiments. Of particular interest are the following observations. Figure 2 (a)
shows that, with only one exception, all quantities start below the equilibrium levels of
the general equilibrium prediction (straight lines in the figures). Most of these variables,
however, exhibit some converge towards these levels. Regarding the development of
prices, panel (b) of Figure 2 exhibits no clear picture concerning the starting levels of
the output prices, but shows that two of the three input prices (r and wl) clearly start
(and seem to stay) below the theoretically predicted levels.
This general impression is confirmed by a convergence analysis based on the follow-
ing estimation model (see Noussair, Plott, and Riezman (1995)):
yit = a11DA1(1/t) + a12DA2(1/t) + a13DA3(1/t) + a2DA(t − 1)/t + uit
where y stands for the particular outcome focused at (quantity, price; with average
outcomes per period and session as units of observation ), i denotes the experimental
0
50
100
150
200
250
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Un
its
(in
tern
atio
nal
an
d la
rge)
0
20
40
60
80
100
120
Un
its
(sm
all)
L_l L_l (theo.) KK (theo.) X X (theo.)Y_l Y_l (theo.) L_sL_s (theo.) Y_s Y_s (theo.)
(a) Quantities
0
0.05
0.1
0.15
0.2
0.25
0.3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Wag
es a
nd
ou
tpu
t p
rice
s
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07C
apit
al p
rice
w_s w_l w (theo.)p_x p_x (theo.) p_ysp_yl p_y (theo.) rr (theo.)
(b) Relative prices
Figure 2 – Development of quantities and prices under the WT-system
16
0
50
100
150
200
250
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Un
its
(in
tern
atio
nal
an
d la
rge)
0
20
40
60
80
100
120
Un
its
(sm
all)
L_l K K (theo.)X Y_l L_sY_s
(a) Quantities
0
0.05
0.1
0.15
0.2
0.25
0.3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Wag
es a
nd
ou
tpu
t p
rice
s
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
Cap
ital
pri
ce
w_s w_l p_x p_ys p_yl r
(b) Relative prices
Figure 3 – Development of quantities and prices under the STLS-system
session, t the trading period in the session, DAi a dummy variable for session i of the
WT-system which is equal to 1 for i and 0 otherwise, and u the error term. Note that
the coefficients a1i indicate session specific starting values and a2 the asymptotic value
of y in the WT-system (DA = 1 when the WT-system is effective). Strong convergence
is said to hold if the estimated asymptotic value (a2) is not significantly different from
the theoretically predicted level. We will speak of weak convergence if the majority of
the starting values (a1i) are further apart from the theoretical level than the estimated
asymptotic value.
The regression results are presented in Table 4. They show that strong convergence
has to be rejected for a majority of the 18 investigated variables. Only the behavior
of employed capital K, employed labor in the small country Ls, and the relative price
pyl satisfy the strong convergence criterion. However, most variables (10 out of 18)
converge in the weak sense. Furthermore, though most asymptotic values are statisti-
cally significantly different from the predicted levels, the differences are mostly small in
economic terms. Given the complexity of the laboratory economy and the fact that the
theoretical general equilibrium model is a very stylized representation of the economy
we find this a quite remarkable result. In line with the visual impression from Figure 2,
we find that the asymptotic value of all aggregate quantity variables is lower than the
respective theoretical level, with the exception of capital employment and the produc-
17
Table 4 – Convergence regressions for constant wage tax regime
Comparison with theoretical benchmark model
Variable a11 a12 a13 a2 prediction p-valuea Wald’s χ2
International
K 231.5 209.2 230.7 236.1 240 .204 18131.7(8.92) (5.99) (4.32) (3.04)
X 171.9 129.0 183.6 164.7 177 .000 5561.2(5.97) (8.22) (14.59) (3.43)
r .0141 .0164 .0201 .0169 .0307 .000 375.4(.0021) (.0011) (.0030) (.0013)
px .1764 .2283 .1869 .2001 .1882 .038 5628.6(.0144) (.0039) (.0082) (.0057)
Small country
Ks 17.7 29.5 42.5 35.9 30 .000 2192.7(4.21) (3.98) (2.55) (1.35)
Ls 28.2 24.9 31.8 25.5 28 .084 811.7(2.48) (3.51) (4.58) (1.43)
Xcs 22.7 20.7 26.1 25.5 22 .000 6512.8
(4.32) (1.12) (2.79) (.70)
Xps 16.3 13.3 27.7 25.0 22 .005 1418.4
(2.71) (1.65) (1.92) (1.05)
Ys 15.4 14.1 18.9 14.8 19 .000 1135.4(2.82) (2.98) (1.72) (.87)
ws .1903 .2010 .1748 .1825 .1694 .001 4962.4(.0110) (.0050) (.0033) (.0039)
pys .2277 .2177 .2500 .2390 .2211 .009 2016.5(.0171) (.0110) (.0120) (.0068)
Large country
Kl 212.3 178.9 187.3 202.5 210 .020 11899.6(11.52) (6.92) (5.72) (3.2)
Ll 167.0 175.5 204.6 186.2 197 .041 2700.0(12.13) (11.26) (14.19) (5.27)
Xcl 147.9 108.3 159.1 142.2 155 .000 4360.2
(6.24) (9.26) (12.37) (3.44)
Xp
l 157.1 114.4 159.3 143.3 155 .000 5710.5(3.69) (8.98) (10.92) (2.97)
Yl 89.9 137.1 113.7 120.3 132 .000 11468.0(12.22) (10.97) (3.38) (2.43)
wl .1796 .1426 .1547 .1501 .1694 .000 5419.4(.0053) (.0048) (.0067) (.0029)
pyl .2107 .2059 .2446 .2211 .2211 .985 30222.6(.0181) (.0102) (.0037) (.0024)
Note:a tests the hypothesis that the asymptotic value a2 is equal to the theoretical pre-
diction; two-sided Wald tests. For X superscript c (p) indicates units consumed (produced);
for Y quantities produced are used as units of observation. Standard errors in parentheses;
corrected for session specific heteroskedasticity and AR(1).
18
tion and consumption of the capital intensive commodity in the small country. Also
two of the three input price variables are too low, while for two of the three output
price variables the asymptote is higher than the theoretical value. This leads to our
first result.
Result 1 A majority of the variables exhibits weak convergence towards the theoretical
general equilibrium levels. The quantity and input price variables are typically converg-
ing from below, while the output prices are typically converging from above.
We have also run convergence regressions for the economic performance indicators un-
employment rate, real GDP, consumer earnings, net capital export, and labor intensity
in the Y-sector. The results of these regressions corroborate the above findings. In
both countries, all five performance indicators are weakly converging to the theoreti-
cally predicted equilibrium values. In both countries the unemployment rate exhibits
even strong convergence (from above) as does real GDP (from below) in the small coun-
try and the Y-production intensity in the large country. There is, however, a caveat to
this result. As will be demonstrated below, this rather positive result does not come
for free, but is associated with relatively large budget deficits.
We now turn to a comparison of the two tax systems in the constant tax regime.
Comparing Figure 2(a) with Figure 3(a) shows that economic activity starts at a lower
level in the experimental sessions with the STLS-system. This holds for the employment
of both input factors, and is accompanied by lower input prices. In particular, output
of the exposed sector X is affected, while its product price px exhibits a clear upward
thrust. To put this outcome into perspective, one has to recognize that in these periods
the small country is facing substantial sales taxes, with a tax rate of 65% and 75% on
the price of X and Y (see Table 1). Recall that these tax rates are not taken from
a theoretical model but determined such that on impact the producers of X and Y
would have to bear the same tax burden as observed under the WT-system. Thus, the
initial economic circumstances are not particularly favorable for a comparatively good
performance of the alternative tax system.
Our primary research questions concern the small country. Therefore, in the fol-
lowing we mainly, but not exclusively, focus on the economic performance regarding
the small country under the two different tax regimes. Figures 4-6 illustrate the de-
velopment of the unemployment rate, the budget surplus, and real GDP, for both tax
19
0
0.1
0.2
0.3
0.4
0.5
0.6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Un
emp
loym
ent
rate
WT-system
STLS-system
(a) Small country
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Un
emp
loym
ent
rate
WT-system
STLS-system
(b) Large country
Figure 4 – Development of unemployment rates under the two tax
systems
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Rel
ativ
e su
rplu
s
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
Tax
rat
e
Surplus WT-system Surplus STLS-system
Wage tax WT-system Sales tax on X STLS-system
(a) Small country
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16Period
Rel
ativ
e su
rplu
s
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Tax
rat
e
Surplus WT-system Surplus STLS-system
Wage tax WT-system Wage tax STLS-system
(b) Large country
Figure 5 – Development of budget surplus and tax rates under the two
tax systems
systems (and both countries). Initially, the unemployment rate in the small country
is at a higher level in case of the STLS-system. However, in spite of the high sales
tax rates, there is a clear tendency for this rate to decline over time (Figure 4). This
stays in clear contrast to the development of the unemployment rate under the WT-
system (and the development in the large country, where a wage tax is applied in both
20
10
12
14
16
18
20
22
24
26
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Rea
l GD
P
WT-system
STLS-system
(a) Small country
30
50
70
90
110
130
150
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Period
Rea
l GD
P
WT-system
STLS-system
(b) Large country
Figure 6 – Development of real GDP under the two tax systems
treatments). Under the STLS-system this appears to have a beneficial effect on the
budget surplus of the small country, which substantially increases over the periods (see
Figure 5). Wage taxes, on the other hand, are systematically accompanied by budget
deficits; this holds for the baseline treatment (WT-system, small and large country)
as well as the alternative treatment (large country with wage tax system). A similar
picture emerges from the development of real GDP (see Figure 6). Whereas economic
activity strongly increases in the small country when the sales tax applies, it shows no
clear development, neither in the small country nor in the large country, when the wage
tax system is effective.
These observations are corroborated by a convergence analysis using an extension
of the estimation model presented above.21 Table 5 gives the results. Whereas the
asymptotic estimates for the large country (b2 versus a2) still show the negative effects
of the relatively adverse start in economic activity in these sessions, the outcomes for
the small country are quite different. Compared to the WT-system, we observe a clear
21The estimation model now becomes
yit = a11DA1(1/t) + a12DA2(1/t) + a13DA3(1/t) + a2DA(t − 1)/t +
b11DB1(1/t) + b12DB2(1/t) + b13DB3(1/t) + b2DB(t − 1)/t + uit
where DBi is a dummy variable representing session i of the STLS-system (equal to 1 for i, 0 oth-
erwise); DB = 1 for sessions where the STLS-system applies in the small country, zero otherwise.
The coefficients b1i denote the session specific starting values and b2 the asymptotic value of y in the
STLS-system.
21
Table 5 – Convergence regressions for constant tax regime
Economic performance indicators compared between the tax systems
Variable a11 a12 a13 b11 b12 b13 a2 b2 p-valuea Wald’s χ2
Small country
Unemploy- .3738 .4479 .2927 .5547 .6151 .2683 .4326 .3134 .001 2667.5ment rate (.0552) (.0779) (.1017) (.0371) (.0357) (.0373) (.0317) (.0135)
Budget −.0927 −.0895 −.0070 .0266 −.0566 .1056 −.1409 .2069 .000 124.3surplus (.0676) (.0804) (.1129) (.0538) (.0923) (.0428) (.0345) (.0251)
Real GDP 13.6 14.1 25.1 15.3 14.2 22.3 20.0 21.8 .039 6392.5(2.02) (1.99) (1.54) (1.44) (1.84) (1.22) (.68) (.56)
Consumer 84.0 88.0 107.1 67.4 18.7 104.6 98.4 87.3 .013 17157.8earnings 1 (7.48) (11.97) (3.60) (6.03) (9.66) (4.78) (2.75) (3.52)
Consumer −13.9 −36.2 172.7 108.4 −25.4 331.5 −59.0 549.7 .000 81.35earnings 2 (80.25) (119.38) (167.24) (147.67) (350.22) (120.64) (48.00) (76.94)
Net capital 8.9 1.3 −16.8 2.8 .4 −.9 −8.1 −9.9 .560 86.5export (6.43) (4.75) (2.85) (4.38) (7.21) (3.42) (2.36) (2.15)
Y-production .5018 .5133 .4011 .5607 .5293 .4975 .3625 .3347 .351 2061.5intensity (.0541) (.0527) (.0438) (.0510) (.0763) (.0522) (.0188) (.0231)
Large country
Unemploy- .4699 .4429 .3506 .6928 .6579 .3653 .4088 .5206 .000 6001.3ment rate (.0385) (.0357) (.0451) (.0257) (.0905) (.0672) (.0167) (.0151)
Budget −.2724 −.1578 −.0425 −.5536 −.6708 −.0388 −.1174 −.2175 .000 1339.8surplus (.0467) (.0154) (.0539) (.0470) (.1587) (.0714) (.0108) (.0258)
Real GDP 120.7 127.2 139.1 71.3 79.5 132.4 134.8 108.6 .000 30313.3(8.16) (2.31) (6.01) (5.25) (18.67) (10.87) (1.64) (3.26)
Consumer 197.3 207.6 214.6 176.9 177.4 197.8 208.9 197.2 .001 136271.6earnings 1 (2.41) (1.93) (2.63) (6.06) (11.93) (5.65) (.95) (3.49)
Consumer −1489.6 −1369.3 −219.5 −3865.3 −3344.2 −379.0 −843.3 −1926.2 .000 1814.1earnings 2 (428.87) (122.11) (492.82) (253.75) (940.85) (665.68) (84.61) (150.48)
Net capital −8.9 −1.3 16.8 −2.8 −.4 .9 8.1 9.9 .560 86.54export (6.43) (4.75) (2.85) (4.38) (7.21) (3.42) (2.36) (2.15)
Y-production .3606 .5322 .4025 .5437 .7096 .6178 .4416 .6612 .000 4060.7intensity (.0257) (.0352) (.0230) (.0493) (.0379) (.0371) (.0157) (.0178)
Note:a tests the hypothesis that the asymptotic values a2 and b2 are equal; two-sided Wald tests. Superscript c (p) indicates units
consumed (produced). Standard errors in parentheses; corrected for session specific heteroskedasticity and AR(1). ‘Unemployment rate’
is defined as the amount of unemployed units of labor relative to the total labor force (endowment) in the respective country; ‘Budget
surplus’ denotes the nominal budget surplus relative to nominal GDP (defined as the total nominal value of the produced goods) in the
respective country; the base ‘year’ for calculating ‘Real GDP’ is the first trading period in each session; ‘Consumer earnings 1’ denotes
average earnings of a consumer in points (‘utility’); ‘Consumer earnings 2’ are ‘Consumer earnings 1’ with the per capita budget surplus
added; ‘Net capital export’ is the difference between total capital sold to the other country and total capital bought from the other
country; ‘Y-production intensity’ denotes the total amount of goods produced in the Y-sector relative to the total amount of goods
produced in the respective country.
22
decrease in the unemployment rate and a substantial improvement in the budget bal-
ance. Also real GDP and consumer earnings net of budget surplus show a statistically
significant better outcome under the STLS- than under the WT-system. Whereas for
the former the better performance is economically not large it is dramatically different
for the latter. When not correcting for budget deficits consumer earnings are statisti-
cally significantly larger under the WT-system. Economically, however, the difference
is not large in that case. The remaining two variables are not significantly different for
the two tax systems. There is no shift in production between the sectors (measured
by ‘Y -production intensity’), while net capital export decreases, but not significantly.
These observations lead to our next result.
Result 2 By the end of the constant tax regime, most economic performance indicators
show a significant improvement for the small country under the STLS-system compared
to the WT-system. Only consumer earnings unadjusted for the budget surplus are sig-
nificantly lower under the STLS-system. In the large country, where in both treatments
the wage tax is applied to finance unemployment benefits, no such development is ob-
served.
Note that these outcomes clearly contradict the intuitive hypothesis concerning the
STLS-system presented at the beginning of the previous section. For constant taxes,
the STLS-system does even better than the WT-system. This holds despite the fact
that the exogenously fixed sales tax rates are in the neighborhood of the unfavorable
general equilibrium.
In the following section we present the results of the trading periods where the tax
rates adjusted to the budget surplus in the previous period: the variable tax regime.
This enables us to investigate the robustness of our findings. Especially, we can examine
the economic impact of changes of tax rates in the different tax systems. Additionally,
it also allows us to test for possible convergence of economic activity in the STLS-
system and, hence, to explore whether economic activity coordinates on one of the two
theoretical general equilibria.
3.2 Variable Tax Regime
When the tax rates start to adjust to the budget surplus in the previous trading period
an economic shock occurs. This can be observed from the development of the quantity
23
variables shown in the panels (a) of Figures 2 and 3. From the former it can be seen
that, under the WT-system, all traded quantities in both countries decrease from period
8 to period 9. Under the STLS-system, the quantities traded internationally and in
the large country also decrease, but now the the traded quantities of local goods in the
small country (Ls and Ys) increase when the tax rates begin to adjust (Figure 3 (a)).
In the last constant tax period all economies with wage taxation are confronted with
substantial budget deficits, whereas large surpluses are generated under the sales tax
system in the small country. Therefore, tax rates increase in the former and decrease in
the latter case (see Figure 5). As illustrated by two economic performance indicators in
Figures 4 and 6, in the economies with wage taxation, this triggers a clearly observable
negative economic shock, with increasing unemployment rates and decreasing real GDP.
Because of this shock, the budget balance does not improve in the transition period 9
(see Figure 5). Thereafter, these economies seem to improve somewhat, showing some
convergence towards a balanced budget and a full utilization of capital (see Figure 3).
However, unemployment stays at a high level, which has a negative effect on outputs,
as manifested by the development of real GDP in Figure 6.22
These developments in the economies where the wage tax system applies are in stark
contrast to the economic development in the small country under the alternative tax
system. First of all, the initial decline in the sales tax rates in period 9 produces positive
economic effects. This is witnessed by the development of the economic performance
indicators in Figures 4 and 6. The unemployment rate drops significantly and real
GDP clearly increases. Note, furthermore, the positive effect on the wage rate (ws),
and the negative effect on the price of the labor intensive good Y (pys), in contrast to the
development under wage taxation (see Figures 2 (b) and 3 (b)). This development is due
to the replacement of the wage tax by a labor subsidy. Remarkably, under the STLS-
system the budget immediately balances, and stays that way over the remaining periods,
with only small deviations. As Figures 4 (a) and 6 (a) indicate, the unemployment rate
and real GDP further improve in later periods, and show convergence towards a level
that is substantially different from the level reached under the WT-system.
22Note, furthermore, that the gap between the values of the economic performance indicators in the
large country narrows over the periods with variable tax rates. We will return to this when presenting
the convergence analysis for the variable tax regime.
24
Table 6 – Convergence regressions for variable tax regime
Economic performance indicators compared between the tax systems
Variable a11 a12 a13 b11 b12 b13 a2 b2 p-valueb Wald’s χ2
Small country
Unemploy- .6647 .5279 .2076 .3200 .2726 .2974 .4807 .2417 .000 2818.1ment rate (.0790) (.0155) (.0845) (.0592) (.0270) (.0411) (.0279) (.0151)
Budget −.3990 −.0357 −.0155 .0151 −.0295 .0061 −.0259 .0038 .210 30.1surplus (.0940) (.0349) (.0450) (.0119) (.0261) (.0177) (.0227) (.0067)
Real GDP 15.1 16.1 26.1 21.3 24.1 22.4 18.9 22.8 .000 8526.7(3.43) (.25) (2.46) (1.76) (1.29) (1.11) (.91) (.56)
Consumer 83.4 86.3 70.7 82.3 76.8 104.8 91.8 89.3 .640 5030.7earnings 1 (4.41) (5.00) (10.25) (8.10) (11.90) (10.82) (2.32) (4.83)
Consumer −353.9 49.6 64.7 122.8 −1.1 114.0 64.2 104.6 .230 447.5earnings 2 (108.24) (47.05) (73.52) (22.09) (64.71) (34.69) (30.73) (13.86)
Net capital 2.0 5.0 −22.0 .1 .7 −23.9 −5.9 −20.1 .000 550.9export (7.48) (1.48) (7.16) (6.47) (3.80) (3.05) (1.04) (1.57)
Y-production .3046 .4229 .4113 .4361 .5061 .4846 .4091 .4204 .709 11007.1intensity (.0471) (.0111) (.0296) (.0654) (.0970) (.0586) (.0107) (.0283)
Large country
Unemploy- .6007 .5046 .4126 .6807 .8258 .6193 .5244 .5838 .043 9030.9ment rate (.0174) (.0676) (.0609) (.0706) (.0842) (.0222) (.0203) (.0213)
Budget −.3274 −.0705 −.0126 −.3562 −.7059 −.3281 −.1249 −.1216 .945 283.4surplus (.0435) (.0705) (.0609) (.1140) 4(.1534) (.0470) (.0344) (.0339)
Real GDP 96.3 117.6 133.3 73.5 48.3 85.1 110.5 98.8 .053 5736.4(3.93) (11.53) (12.01) (15.34) (18.14) (4.99) (3.91) (4.61)
Consumer 196.7 202.4 211.1 188.7 169.5 192.4 203.9 197.5 .008 101716.7earnings 1 (2.16) (6.74) (3.35) (3.19) (8.90) (3.88) (1.51) (1.91)
Consumer −2343.8 −550.1 93.3 −2155.2 −4374.2 −1830.1 −750.2 −935.9 .631 222.7earnings 2 (366.52) (579.34) (481.54) (948.01) (1154.18) (323.99) (282.92) (264.15)
Net capital −2.0 −5.0 22.0 −.1 −.7 23.9 5.9 20.1 .000 550.9export (7.48) (1.48) (7.16) (6.47) (3.80) (3.05) (1.04) (1.574)
Y-production .4409 .4735 .4671 .6344 .6051 .5444 .4484 .5411 .002 32879.1intensity (.0122) (.0528) (.0107) (.0409) (.0860) (.0539) (.0071) (.0289)
Note:a tests the hypothesis that the asymptotic values a2 and b2 are equal; two-sided Wald tests. Superscript c (p) indicates units
consumed (produced). Standard errors in parentheses; corrected for session specific heteroskedasticity and AR(1). ‘Unemployment rate’
is defined as the amount of unemployed units of labor relative to the total labor force (endowment) in the respective country; ‘Budget
surplus’ denotes the nominal budget surplus relative to nominal GDP (defined as the total nominal value of the produced goods) in the
respective country; the base ‘year’ for calculating ‘Real GDP’ is the first trading period in each session; ‘Consumer earnings 1’ denotes
average earnings of a consumer in points (‘utility’); ‘Consumer earnings 2’ are ‘Consumer earnings 1’ with the per capita budget surplus
added; ‘Net capital export’ is the difference between total capital sold to the other country and total capital bought from the other country;
‘Y-production intensity’ denotes the total amount of goods produced in the Y-sector relative to the total amount of goods produced in the
respective country.
25
Table 6 presents the results of the convergence analysis comparing the performance
of the two tax systems for the variable tax regime. These estimation results corroborate
the above observations.
Comparing the estimated asymptotic values a2 and b2, for the small country under
the STLS-system, a significant decrease in the unemployment rate and net capital
export together with a significant increase in real GDP show up. For the budget
surplus, the labor intensity of production, and both of the consumer earnings measures,
no statistically significant differences are found. Observe, however, that the amount of
consumer earnings adjusted for the budget surplus shows a considerable improvement,
too. The outcome of no significant difference in the development of the budget surplus
is due to the convergence towards a balanced budget under both tax systems when tax
rates adjust. Notice, however, that the convergence happens from a negative balance
under the WT-system whereas it convergences from a surplus under the STLS-system.
Not surprisingly, for the large country, the outcomes are worse for the STLS-system
sessions, because of the bad start. Note, however, that the asymptotic values a2 and b2,
which are statistically significantly different for unemployment and consumer earnings
uncorrected for the budget surplus, clearly show a movement towards each other. For
both indicators the starting values are much further apart than the asymptotic values.
Furthermore, the differences seem economically not significant. This pattern is in line
with the observation from the figures indicating that the gap between the values of
the economic performance indicators for this country narrows over the periods with
variable tax rates. The budget surplus is clearly negative and virtually the same under
both systems as are the consumer earnings net of the budget surplus. The significant
difference in net capital export mirrors the result for the small country. The following
result summarizes.
Result 3 Under the variable tax regime, the positive view of the STLS-system as ob-
served for constant taxes is corroborated and enhanced. The economic performance of
the country where the STLS-system is applied further improves and shows a substan-
tially lower unemployment rate and net capital export, as well as a higher real GDP,
compared to its performance under the WT-system. With respect to the other economic
indicators - the budget surplus, consumer earnings, and labor intensity of production -
there are no significant differences in performance.
26
An important further issue concerns the economic effect of changes in the different
tax rates under the two tax systems. Table 7 shows the results of a regression analysis
with respect to the economic performance indicators: unemployment rate, capital em-
ployment, real GDP, consumer earnings, net capital flight, and Y -production intensity.
In addition to the tax rates the number of the trading period is also included as an
explanatory variable, to control for a time trend.
Several observations are in order. First of all, the signs of all tax effects are com-
pletely in line with economic intuition. For both the wage tax and the sales tax it
appears that tax hikes have a negative impact on economic activity and consumer
earnings of the country directly involved. Higher taxes also encourage capital flight.
Furthermore, changes of tax rates in the small country have no spill-over effects on the
large country (the only exception being the effect of a wage tax change on capital flight,
which is due to definition of this variable). An increase of the wage tax in the large
country, however, has a statistically significantly negative effect on consumer earnings
in the small country. The only obscure result concerns the effect of the sales tax on
labor intensity of production.
The regression results clearly show that a wage tax increase has strong adverse
effects on the economic performance in the respective country. This is witnessed by
the statistically and economically highly significant coefficients of the wage taxes τws
in the small country and τwl in the large country, in most regressions. Increasing the
wage tax rate in a country substantially increases unemployment and capital flight and
decreases real GDP and capital employment.
An increase of the sales tax rate in the small country also adversely affects unem-
ployment and real GDP in a statistically significant way. What is striking, though, is
that the magnitude of these effects is substantially smaller than the effects of a wage
tax rate increase. For unemployment the coefficient is .4420 for the the wage tax but
only .1831 for the sales tax. Similarly, real GDP decreases by only 6.85 when the sales
tax increases whereas the marginal decrease of this measure amounts to 11.95 for the
wage tax. For capital employment, consumer earnings, net capital flight, and labor
intensity of production, a change in the sales tax is not even significantly different from
zero. The next result summarizes the most important findings.
27
Table 7 – The effect of taxes on the performance of
real economic variables
Unemploy- Capital Consumer Net capital Y-production
Variable ment rate employment Real GDP earnings 1 flight intensity
Small country
τws .4420∗∗∗−27.95∗∗∗
−11.95∗∗∗−2.89 24.42∗∗∗
−.0879(.0554) (6.02) (1.92) (8.18) (5.58) (.0521)
τxs .1831∗−8.87 −6.85∗∗∗
−3.38 13.96 −.0639(.0728) (7.10) (2.03) (13.22) (7.95) (.0750)
τwl −.0342 3.82 −.25 −22.21∗−2.55 .0308
(.0425) (5.70) (1.32) (9.74) (5.95) (.0604)
period −.01 1.11 −.03 1.15∗−1.44∗∗∗ .0018
(.05) (.54)∗ (.14) (.56) (.38) (.0038)
constant .26∗∗∗ 31.30∗∗∗ 26.72∗∗∗ 95.90∗∗∗−.25 .4136∗∗∗
(.06) (6.24) (1.78) (6.44) (4.37) (.0430)
N 54 54 54 54 54 54
Wald’s χ2 106.2 25.5 42.1 16.6 48.0 5.1
Large country
τws .0041 32.95 5.69 6.30 −24.42∗∗∗−.0357
(.0528) (18.29) (11.31) (5.00) (5.58) (.0628)
τxs .1450 −26.03 −27.46 −8.61 −13.96 .1964∗∗∗
(.0808) (25.40) (17.35) (8.40) (7.95) (.0627)
τwl .2793∗∗∗−49.81∗
−54.71∗∗∗−25.61∗∗∗ 2.55 .0238
(.0583) (20.27) (12.76) (5.39) (5.95) (.0472)
period −.0032 .4667 .4183 .6010 1.44∗∗∗ .0005(.0043) (1.18) (1.00) (.4271) (.38) (.0032)
constant .3371∗∗∗ 203.58∗∗∗ 147.29∗∗∗ 210.60∗∗∗ .25 .4485∗∗∗
(.0487) (13.06) (11.39) (38.46) (4.37) (.0367)
N 54 54 54 54 54 54
Wald’s χ2 38.2 54.1 27.4 24.6 48.0 19.0
Note: ∗∗∗ significant at .1 percent, ∗∗ significant at 1 percent, and ∗ significant at 5 percent. Standard
errors in parentheses; corrected for session specific heteroskedasticity and AR(1). All estimates are based
on periods 8 to 16. ‘Unemployment rate’ is defined as the amount of unemployed units of labor relative
to the total labor force (endowment) in the respective country; ‘Capital employment’ denotes the total
amount of capital employed in the respective country; ‘Real GDP’ is total value of produced goods in the
respective country with the first trading period in each session as the base ‘year’; ‘Consumer earnings 1’
denotes the average earning of a consumer in points (‘utility’); ‘Net capital flight’ is the difference between
total capital sold to the other country and total capital bought from the other country; ‘Y-production
intensity’ denotes the total amount of goods produced in the Y-sector relative to the total amount of
goods produced in the respective country. Only one sales tax rate appears in the regressions because of
the fixed ratio of the tax rates for the two production sectors (see Table 1).
28
Result 4 In the small country, compared to the wage tax, an increase in the sales tax
appears to have a much smaller adverse economic impact. The detrimental effect of a
rise in the sales tax on the employment of labor and real GDP is only about half the
effect of an increase in the wage tax. Moreover, whereas there is a significantly negative
effect of an increase in the wage tax on capital, no such effect is found for the sales tax.
All in all, compared to the WT-system, the performance of the STLS-system turns out
to be remarkably good.
What remains to be discussed is whether under the STLS-system economic activities
tend to coordinate on one of the two theoretical general equilibria presented in Table 2.
To this end we performed a convergence analysis for the periods 9 to 16 and compared
the outcomes with the theoretical predictions. Table 8 presents the results. In the
table, column “equil. 1” shows the predictions of the ‘favorable’ general equilibrium
and the column labeled “equil. 2” those of the ‘unfavorable’ one. For the international
variables, which are mainly influenced by the large country, it holds that too little
capital is employed and (accordingly) too little X produced. Both variables, however,
exhibit weak convergence towards the equilibrium values which are virtually identical
in the two general equilibria. On the other hand, the price of capital is too low and the
price of commodity X appears to be too high. All this suggests that the equilibration
process did not yet settle down on the international markets. A similar picture can
be observed for the development of variables in the large country. There, capital and
labor employment as well as the production and consumption of X weakly converge
towards the equilibrium values from below. Concerning the local prices of labor and Y
it can be observed that the former is too low and the latter to high compared to any
of the equilibrium predictions. The budget surplus is weakly balancing from below but
the tax rate τwl is much too high and far removed from any equilibrium value.
We turn now to the most interesting case, the small country in which the alternative
STLS-system applies. Here the results are remarkably different. Three of the five real
variables converge at least weakly to the favorable equilibrium (Ls, Xcs , and Ys). Capital
K and the production level of the capital intensive commodity, Xps , do not converge
but do better than predicted in the favorable equilibrium. The budget balance strongly
converges to zero, while the sales taxes (weakly) converge to the tax rates predicted
by the favorable general equilibrium. Also the wage rate converges from below to the
value predicted in the favorable equilibrium. Only the price of the local good Y does
29
Table 8 – Convergence regressions for variable STLS regime
Comparison with theoretical equilibrium predictions
Variable b11 b12 b13 b2 equil. 1 p-valuea equil. 2 p-valuea Wald’s χ2
International
K 188.4 131.6 201.7 209.3 240 .000 240 .000 5050.2(23.73) (10.39) (12.37) (4.73)
X 82.0 91.7 95.0 132.1 181 .000 182 .000 473.7(21.0) (21.4) (12.3) (7.86)
r .0036 .0010 .0011 .0006 .0295 .000 .0289 .000 250.2(.0003) (.0003) (.0006) (.0002)
px .2313 .3450 .2191 .2427 .1807 .000 .1807 .000 2624.9(0.35) (0.97) (0.44) (0.32)
Small country
Ks 29.8 36.6 50.4 51.1 28 .000 11 .000 451.1(7.69) (7.17) (4.32) (3.63)
Ls 30.6 32.7 31.6 34.1 33 .098 18 .000 5719.6(2.67) (1, 21) (1.85) (.68)
Xcs 11.9 9.6 19.0 18.9 25 .000 17 .235 260.7
(2.76) (5.42) (4.30) (1.63)
Xps 27.5 23.8 22.4 30.0 25 .001 14 .000 1038.2
(4.03) (5.55) (2.21) (1.44)
Ys 20.5 23.6 21.4 21.2 21 .917 11 .000 370.6(4.13) (4.01) (3.32) (1.60)
ws .1566 .1099 .1998 .1647 .1971 .011 .1292 .005 237.0(.0144) (.0178) (.0235) (.0128)
pys .2074 .0979 .1950 .1704 .2165 .000 .2747 .000 13928.1(.0048) (.0309) (.0329) (.0043)
τpxs .5442 .2985 .5239 .4681 .4889 .035 .7835 .000 8102.9(.0143) (.0286) (.0264) (.0099)
surs .0151 −.0295 .0061 .0038 0 .568 0 .568 8086.3(.0119) (.0261) (.0177) (.0067)
Large country
Kl 160.3 105.5 148.3 168.6 212 .000 229 .000 1809.9(17.84) (13.77) (17.14) (5.96)
Ll 100.6 54.9 119.9 131.1 199 .000 213 .000 1749.2(22.24) (26.53) (7.00) (6.70)
Xcl
69.5 82.1 75.6 111.5 156 .000 165 .000 488.2(14.26) (18.60) (10.86) (6.42)
Xpl
54.2 61.6 74.5 100.4 156 .000 168 .000 300.5(19.98) (18.53) (11.68) (7.51)
Yl 100.4 32.9 99.7 92.3 133 .000 140 .000 1832.8(10.24) (9.57) (6.97) (3.49)
wl .1451 .1533 .1538 .1446 .1640 .000 .1743 .000 4289.4(.0037) (.0052) (.0065) (.0032)
pyl .2380 .3103 .2313 .2512 .2123 .000 .2121 .000 1934.6(.0069) (.0217) (.0183) (.0111)
τwl .8762 .8982 .8258 .8958 .3655 .000 .2769 .000 141093.6(.0418) (.0067) (.0228) (.0156)
surl −.3562 −.7059 −.3281 −.1216 0 .000 0 .000 207.5(.1140) (.1534) (.0470) (.0339)
Note:a tests the hypothesis that the asymptotic value a2 is equal to the theoretical prediction; two-sided
Wald tests. For X superscript c (p) indicates units consumed (produced); for Y quantities produced are
used as units of observation. Standard errors in parentheses; corrected for session specific heteroskedasticity
and AR(1). Only the sales tax rate τpxs appears because of the exogenously fixed ratio of the tax rates for
the two production sectors.
30
not show a clear pattern in the small country. The following result summarizes the
main observations.
Result 5 Under the variable tax regime, when the STLS-system obtains in the small
country and the WT-system obtains in the large country, the international real variables
and the real variables related to the large country mostly weakly converge towards to the
equilibrium predictions, from below. The input prices turn out to be too low and the
output prices too high compared to both theoretical equilibria. In contrast, in the small
country the real variables are either converging towards the favorable equilibrium or do
even better than this equilibrium predicts. The wage rate also converges towards the
prediction of the favorable equilibrium. Only the output price for the labor intensive
commodity Y stays too low compared to any of the equilibrium predictions.
Hence, the alternative STLS-system performs better than the WT-system by coordi-
nating economic activity in the direction of the favorable general equilibrium or does
even better than that. In the next section we offer a tentative explanation for this
finding, using theoretical arguments and some further evidence.
4 A behavioral explanation and empirical support
Our experimental results clearly suggest that financing unemployment benefits via sales
taxes, in combination with a subsidy for employment, leads to much better economic
outcomes than using a wage tax, even in a relatively small open economy. Though
disadvantaged at the beginning, due to high tax rates stemming from persistent budget
deficits under the wage tax system and the requirement of equivalent tax burdens
on impact, the STLS-system manifests its beneficial effects immediately. Its better
performance regarding the economic indicators is even further improved when tax rates
start to adjust to the budget balance. Also, it appears that changes in sales taxes have
a much weaker negative economic effect than changes in wage taxes.
Compared to the theoretical predictions of the standard general equilibrium model
the wage tax system has a tendency to do worse. For the STLS-system the theoretical
prediction is ambiguous in the sense that two stable equilibria exist. It turns out that
under this alternative system economic behavior exhibits a tendency to coordinate
on the ‘favorable’ equilibrium or to perform even better than the theoretical model
predicts. In the following we offer theoretical arguments and empirical support for the
following claim, which can explain these observations.
31
Claim 1 Uncertainty about product prices makes producers reluctant to incur produc-
tion costs. This can explain the good economic performance of the sales-tax-cum-labor-
subsidy system in comparison with the wage tax system. Instead of being confronted
with a tax burden up-front on the input of labor, producers under the former system
receive a labor subsidy and only have to pay taxes in proportion to their sales revenues,
which effectively means risk sharing by the government.
To substantiate this claim we offer four pieces of evidence. First, recall from Result 1
that under the constant wage tax regime quantity and input price variables typically
converge from below, whereas output prices seem to converge from above towards the
competitive equilibrium levels of the theoretical general equilibrium model. Although
this theoretical model does not capture all details of our complex lab economy, the
result is suggestive of some downward pressure on the demand for inputs. Also, because
these outcomes are accompanied by a budget deficit. A second piece of evidence in this
respect is obtained by comparing the (after tax) marginal revenue product of labor and
capital with the respective net (i.e. after tax or subsidy) input price. Table 9 shows the
number of cases in which producers’ marginal revenue product exceeds the input price,
using average current prices.23 Assuming random errors, profit maximization would be
consistent with a fraction of 50%. The observed fractions are remarkably different from
this benchmark, however. Our next result summarizes the evidence.
Result 6 Averaging over periods and tax systems, for about 70 percent of the cases
producers’ marginal revenue product of capital (74%) and labor (65%) exceeds the input
price. Taken over all periods, with only one exception in each system, the excess is
always significant. Moreover, comparing the second half of the trading periods with the
first half, there is no systematic decrease in the excess.
This result provides further support for the view that, under both tax systems, produc-
ers are reluctant to buy inputs.24 Interestingly, Noussair, Plott, and Riezman (1995)
observe a similar phenomenon in an experiment concerning international trade.25 These
23Similar results are obtained when the average product price of the previous period is used.24In particular, because we have no evidence of a shortage of capital or labor. On the contrary,
comparing actual labor supply with theoretical labor supply - using the benchmark model and actual
prices - we find excess supply for a fraction of consumers that is significantly larger than 50% (on
average, 94% for the WT-system and 73% for the STLS-system). Concerning capital, the relatively
low capital price also points into the direction of an excess supply (see Figures 2 and 3).25Noussair, Plott, and Riezman (1995) use simultaneous (double auction) input and output markets.
Thus, our results do not seem to be due to the use of sequential markets. Hey and di Cagno (1998),
investigating experimentally two sequential double auction markets, also observe a shortage of trade,
compared to the competitive equilibrium predictions (see also Goeree and Holt 1999).
32
Table 9 – Fraction of cases where producers’marginal revenue product exceeds net input price
WT-system STLS-system
Labor Capital Labor Capital
Small country
1 to 8 0.4583 0.6250 0.6250 0.9583
(0.7646) (0.0557) (0.0557) (0.0000)
X-sector 9 to 16 0.5000 0.6667 0.8750 0.9583
(0.5573) (0.0147) (0.0000) (0.0000)
all 0.4792 0.6458 0.7500 0.9583
(0.6950) (0.0028) (0.0000) (0.0000)
1 to 8 0.5694 0.6944 0.4722 0.6528
(0.1444) (0.0007) (0.7220) (0.0064)
Y-sector 9 to 16 0.5833 0.8472 0.9167 0.6944
(0.0973) (0.0000) (0.0000) (0.0007)
all 0.5764 0.7708 0.6944 0.6736
(0.0399) (0.0000) (0.0000) (0.0000)
Large country
1 to 8 0.6042 0.8333 0.9583 0.9583
(0.0967) (0.0000) (0.0000) (0.0000)
X-sector 9 to 16 0.6250 0.7708 0.7708 1.0000
(0.0557) (0.0001) (0.0001) (0.0000)
all 0.6146 0.8021 0.8646 0.9792
(0.0158) (0.0000) (0.0000) (0.0000)
1 to 8 0.6250 0.6528 0.7361 0.5694
(0.0222) (0.0064) (0.0000) (0.1444)
Y-sector 9 to 16 0.6111 0.5972 0.5556 0.6944
(0.0382) (0.0625) (0.2048) (0.0007)
all 0.6181 0.6250 0.6458 0.6319
(0.0029) (0.0017) (0.0003) (0.6319)
Note: Based on average current period input and output prices and all
periods; within parentheses the probability of obtaining values as least
as extreme as observed when p = 0.5; binomial test, one-sided; n = 48
for periods 1 to 8 and 9 to 16, n = 96 for all periods.
authors conjecture that producers may require a compensation for the market risk they
run, since they may not be able to sell outputs. A likely underlying reason for this
behavior is some form of aversion towards risk or losses. Indeed, theoretical partial
equilibrium models exist indicating that product price uncertainty reduces the factor
demand of risk-averse competitive firms (see the literature mentioned in the Introduc-
33
Table 10 – Correlation of factor employment and
output price uncertainty
St.dev. output price St.dev. output price
WT-system STLS-system WT-system STLS-system
Lys -0.0868 -0.3490 Kys -0.0056 -0.3390(0.5709) (0.0188) (0.9711) (0.0227)
Lyl -0.3198 -0.4796 Kyl -0.3251 0.2455(0.0322) (0.0009) (0.0293) (0.1041)
Lxs -0.3195 -0.5016 Kxs -0.3004 -0.6194(0.0324) (0.0004) (0.0450) (0.0000)
Lxl -0.2721 -0.6669 Kxl 0.0737 -0.5833(0.0706) (0.0000) (0.6304) (0.0000)
Lx -0.2828 -0.6565 Kx -0.0523 -0.6567(0.0598) (0.0000) (0.7331) (0.0000)
Note: Entries show Spearman’s ρ between employment of the mentioned factor in
period t and the standard deviation of the relevant nominal output price in period
t − 1; p-values in parentheses, two-sided tests.
tion), and risk-averse behavior of firms appears to be a realistic assumption.26 Yet,
empirical microeconomic studies of the consequences of market uncertainty for factor
demand are scarce (see Ghosal (1995)).27 Important exceptions are Leahy and Whited
(1996) and Guiso and Parigi (1999). Both of these studies find that investment is
negatively affected by uncertainty.
Taking the standard deviation of transaction prices in the previous period as mea-
sure of expected price uncertainty in the current period, we examine the correlation of
this measure with the employment of capital and labor. Table 10 presents the outcomes.
In line with our conjecture and as suggested by field studies, a mostly significantly neg-
ative correlation shows up. The next result summarizes this third piece of evidence for
our claim.
Result 7 The demand for capital and labor is mostly significantly negatively correlated
with output price uncertainty.
What causes risk averse behavior is not completely clear yet. Recent studies on the
economic significance of emotions suggest that anxiety may play a role, because of the
26According to Stiglitz (1999, p. 254): “There is by now a large body of literature arguing that
normally firms act in a risk averse manner (...)”. Zhang (1998, p. 1753) notes: “Investors of all types
generally exhibit aversion to risk”. For an empirical study showing risk-aversion by firms, see Gunjal
and Legault (1995).27The situation is different for studies focusing on the impact of macroeconomic uncertainty, as
captured by inflation, for instance. These studies typically show a negative effect on private investment
(see Aizenman and Marion 1993, Brunetti and Weder 1998).
34
time lag between inputs and outputs (cf. the motto of our paper). For example, Caplin
and Leahy (2001) argue that by ignoring anxiety conventional measures of risk aversion
underestimate the effects of uncertainty on asset prices. The reason is that because
anxiety is aversive it requires compensation by a higher rate of return. Consequently, an
anxious decision maker may appear more risk averse.28 Another relevant finding in this
context is that the possibility rather than the probability of a negative outcome appears
to be important (Loewenstein, Hsee, Weber, and Welch (2001)), which manifests itself
in overreaction to small probability events (see Harless and Camerer 1994). In the
context of sequential markets it is also interesting that people also seem to treat delayed
outcomes as being uncertain (see Keren and Roelofsma (1995)). These results do not
only provide additional support for our result of a negative effect of price uncertainty
on factor demand, they also indicate that little perceived uncertainty may already
have substantial effects. Thus, it need not be surprising if we do not observe rapid
convergence to competitive equilibrium levels in complex market environments. To
improve theoretical predictions it seems important to take the so far neglected dynamic
behavioral aspects of such market economies into account. A research direction which
is strongly advocated by Akerlof (2002).
Importantly, the overall output price uncertainty turns out to be similar under both
tax systems. In fact, our measure of uncertainty shows some tendency to be larger under
the STLS-system. Together with the above result and the relatively good performance
of the economy under this system this corroborates the view that producers perceive the
uncertainty differently under the WT- and the STLS-system. The fact that under the
latter system the risk can be shared with the government seems to play an important
role here.
For our fourth and final piece of evidence for our claim we return to Table 7. This
table shows that increases in the sales tax have much weaker adverse economic effects
than increases in the wage tax. This finding fits the view that producers are relatively
more concerned with incurring certain costs up-front than with some uncertain costs,
that can be shared with the government, in the future.
All in all, the theoretical arguments and the empirical support for our claim seem
substantial. The evidence presented makes it quite intuitive why the sales tax system
28Experimental evidence of a negative impact of anxiety on risk taking is presented in Bosman and
van Winden (2002).
35
performs so much better than the wage tax system. It also provides a rationale for why
under the STLS-system economic activity is attracted by the favorable equilibrium or
does even better than that. With any aversion to risk or losses, having the government
sharing in the sales revenue risk, instead of having to bear a tax burden up-front,
certainly seems to be the more producer and employment friendly scheme. The more
so when a labor subsidy is included.
5 Conclusion
In this paper we present an experimental comparison of a wage tax system and an
alternative sales-tax-cum-labor-subsidy system as a means of financing unemployment
benefits. Our experimental results are strongly in favor of the alternative system. Un-
der the alternative system, economic activity is directed towards the theoretical equi-
librium without capital flight and with the better economic performance. In particular,
employment and GDP do not decline but sharply increase. Instead of capital flight,
even capital import is observed and the balance of the budget does not deteriorate but
strongly improves.
In our view, producers’ reluctance to incur production costs up-front when facing
product price uncertainty plays a crucial role. We present four pieces of experimental
evidence in support of this claim. First, a convergence analysis using the results of the
theoretical benchmark model appears to be suggestive of a downward pressure on the
demand for inputs. Quantity and input price variables typically converge from below,
and output price variables from above the theoretically predicted levels. Second, for
most producers the estimated marginal revenue product of capital and labor persistently
exceeds the respective input price. Third, the demand for capital and labor turns out
to be mostly negatively correlated with the variance of output transaction prices in
the previous trading period. Finally, we find that increases in sales tax rates have
much weaker adverse economic effects than increases in wage tax rates. This evidence
makes it understandable that the alternative tax system performs much better than
the wage tax system. Instead of having to pay an input tax up-front, producers receive
a labor subsidy while they only have to pay taxes in proportion to whatever the sales
revenues turn out to be. The latter effectively means risk sharing by the government.
36
Furthermore, our claim finds support from some theoretical partial equilibrium models
showing that risk-averse firms indeed employ fewer inputs.
Although, as yet, relatively few macroeconomic experiments have been carried out,
we think that the experience that is now accumulating is of interest from a scientific
as well as policy perspective. For example, our experimental findings are in agreement
with Akerlof’s view that macroeconomics should be behavioral, in the spirit of Keynes
(see Akerlof (2002, p. 428)). An aversion on the side of producers to input taxation
relative to output taxation, due to the elapse of time, is a behavioral factor that seems
to have been neglected in theory. In fact, this finding may have a wider bearing on
the theoretical modeling of how economic agents behave in complex dynamic market
environments. As noted by Plott (2001): “as it turns out, the classical theories of price
adjustment are incomplete” (p. 3), and “experiments teach us about theory and it is
theory that we use when addressing complex and new problems. The progress builds
in slow and in unexpected ways” (p. 27).
37
References
Aizenman, J., and N. P. Marion (1993): “Macroeconomic uncertainty and private
investment,” Economics Letters, 41, 207-210.
Akerlof, G. A. (2002): “Behavioral Macroeconomics and Macroeconomic Behavior,”
American Economic Review, 92, 3, 411-433.
Basu, S. and J. G. Fernald (1997): “Returns to scale in U.S. production: estimates
and implications,” Journal of Political Economy, 105, 2, 249-283.
Batra, R. N., and A. Ullah (1974): “Competitive firm and the theory of input
demand under price uncertainty,” Journal of Political Economy, 82, 3, 537-548.
Blanchard, O., and L. F. Katz (1997): “What We Know and Do Not Know About
the Natural Rate of Unemployment,” Journal of Economic Perspectives, 11, 1,
51-72.
Bosman, R., and F. van Winden (2002): “Global risk, effort, and emotions in an
investment experiment,” CREED working paper, University of Amsterdam.
Bovenberg, A. L. (1994): “Perspectives on tax policy in small and open economies,”
Scandinavian Journal of Economics, 96, 3, 283-287.
Braulke, M., and G. Corneo (2003): “Capital taxation may survive in open
economies,” CESifo Working Paper No.975. University of Munich.
Brunetti, A., and B. Weder (1998): “Investment and institutional uncertainty: a
comparative study of different uncertainty measures,” Weltwirtschaftliches Archiv,
134, 3, 513-533.
Caplin, A., and J. Leahy (2001): Psychological expected utility theory and antici-
patory feelings,” Quarterly Journal of Economics, CXVI, 55-81.
Davis, D. D., and C. A. Holt (1993): Experimental Economics, Princeton University
Press: Princeton.
de Mooij, R.A., and A.L. Bovenberg (1998): “Environmental taxes, international
capital mobility and inefficient tax systems: Tax burden vs. tax shifting,” Inter-
national Tax and Public Finance, 5, 7-39.
38
Diamond P.A., and J.A. Mirrlees (1971): “Optimal taxation and public production
I: Production efficiency, and II: Tax rules,” American Economic Review, 61, 8-27
and 261-278.
Dierker, E. (1972): “Two remarks on the number of equilibria of an economy,” Econo-
metrica, 40, 951-953.
Eaton, J. and H. S. Rosen (1980): “Labor supply, uncertainty, and efficient taxa-
tion,” Journal of Public Economics, 14, 365-374.
Ghosal, V. (1995): “Input Choices under Price Uncertainty,” Economic Inquiry, 33,
1, 142-158.
Goeree, J. K., and C. A. Holt (1999): “Employment and prices in a simple macroe-
conomy,” Southern Economic Journal, 65, 3, 637-647.
Gordon R. H. (1992): “Can capital income taxes survive in open economies?” The
Journal of Finance, XLVII, 3, 1159-1180.
Guiso, L., and G. Parigi (1999): “Investment and demand uncertainty,” Quarterly
Journal of Economics, CXIV, 185-227.
Gunjal, K., and B. Legault (1995): “Risk Preferences of Dairy and Hog Producers
in Quebec,” Canadian Journal of Agricultural-Economics, 43, 1, 23-35.
Harless, D. W., and C. Camerer (1994): “The predictive utility of generalized
expected utility theories,” Econometrica, 62, 1251-1289.
Hartman, R. (1975): “Competitive firm and the theory of input demand under price
uncertainty: comment,” Journal of Political Economy, 83, 6, 1289-1290.
Hartman, R. (1976): “Factor demand with output price uncertainty,” American Eco-
nomic Review, 66, 4, 675-81.
Hey, J. D. and D. di Cagno (1998): “Sequential markets: an experimental investiga-
tion of Clower’s dual-decision hypothesis,” Experimental Economics, 1, 1, 63-85.
Holthausen, D. M. (1976): “Input choices and uncertain demand,” American Eco-
nomic Review, 66, 1, 94-103.
39
Huck, S., K. A. Konrad, W. Muller, and H. T. Norman, H. (2001): “Mergers
and the perception of market power: An experimental study,” SSRN working
paper series.
Kahneman, D., and A. Tversky (1979): “Prospect Theory: An Analysis of Decision
under Risk ,” Econometrica, 47, 2, 263-291.
Keren, G. and P. Roelofsma (1995): “Immediacy and certainty in intertemporal
choice,” Organizational Behavior and Human Decision Processes, 63, 287-297.
Keynes, J. M. (1970): The General Theory of Employment, Interest and Money.
MacMillan: London.
Koskela, E., and R. Schob (2002): “Why governments should tax mobile capital in
the presence of unemployment,” Contributions to Economic Analysis and Policy,
1, 1, Article 1.
Layard, R., S. Nickell, and R. Jackman (1991): Unemployment: Macroeconomic
Performance and the Labour Market. Oxford University Press: Oxford.
Leahy, J.V., and T.M.. Whited (1996): “The effect of uncertainty on investment:
some stylized facts,” Journal of Money, Credit, and Banking, 28, 1, 64-82.
Lian, P. and C. R. Plott (1998): “General equilibrium, markets, macroeconomics
and money in a laboratory experimental environment,” Economic Theory, 12, 21-
75.
Loewenstein, G. F., C. K. Hsee, E. U. Weber, and N. Welch (2001): “Risk as
feelings,” Psychological Bulletin, 127, 267-286.
Noussair, C. N., C. R. Plott, and R. G. Riezman (1995): “An experimental
investigation of the patterns of international trade,” American Economic Review,
85, 462-491.
Noussair, C. N., C. R. Plott, and R. G. Riezman (1997): “The principles of
exchange rate determination in an international finance experiment,” Journal of
Political Economy, 105, 4, 822-861.
40
Plott, C. (2001): “Equilibrium, equilibration, information and multiple markets: from
basic science to institutional design,” California Institute of Technology, mimeo.
Plott, C. R., and P. Gray (1990): “The multiple unit double auction,” Journal of
Economic Behavior and Organization, 13, 245-258.
Quirmbach, H. C., C. W. Swenson, and C. C. Vines (1996): “An experimental
examination of general equilibrium tax incidence,” Journal of Public Economics,
61, 337-358.
Razin, A., and E. Sadka (1991): “International tax competition and gains from tax
harmonization,” Economics Letters, 37, 69-76.
Richter W. F., and K. Schneider (2001): “Taxing mobile capital with labor market
imperfections,” International Tax and Public Finance, 8, 245-262.
Rogerson, R. (1997). ‘Theory Ahead of Language in the Economics of Unemployment’,
Journal of Economic Perspectives, vol. 11, pp. 73-92.
Roth, A. E. (2002): “The economist as engineer: game theory, experimentation, and
computation as tools for design economics,” Econometrica, 70, 4, 1341-1378.
Sandmo, A. (1971), “On the theory of the competitive firm under price uncertainty,”
The American Economic Review, 61, 65-73.
Snower, D. J. (2000): “Evolution of the Welfare State,” Schriften des Vereins fur
Socialpolitik, 271, 35-52.
Sørenson, P. B. (1997): “Public Finance Solutions to the European unemployment
problem?,” Economic Policy: A European Forum, 25, 221-251.
Stiglitz, J. E. (1999): “Taxation, Public Policy, and Dynamics of Unemployment ”,
International Tax and Public Finance, 6, 3, 239-262.
van Winden, F. A. A. M., A. Riedl, J. Wit, and F. van Dijk (1999): “Exper-
imenteel Onderzoek naar het Plan Van Elswijk,” CREED, University of Amster-
dam.
Zhang, G. (1998): “Ownership concentration, risk aversion, and the effect of financial
structure on investment decisions,” European Economic Review, 42, 1751-1778.
41